metadata

license: apache-2.0
tags:
  - whisper-event
  - generated_from_trainer
datasets:
  - facebook/multilingual_librispeech
metrics:
  - wer
model-index:
  - name: Whisper largeV2 dutch MLS
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: facebook/multilingual_librispeech dutch
          type: facebook/multilingual_librispeech
          config: dutch
          split: test
          args: dutch
        metrics:
          - name: Wer
            type: wer
            value: 10.591602311347534

Whisper largeV2 dutch MLS

This model is a fine-tuned version of openai/whisper-large-v2 on the facebook/multilingual_librispeech dutch dataset. It achieves the following results on the evaluation set:

Loss: 0.2031
Wer: 10.5916

Model description

The model is fine-tuned for 4000 updates/steps on multilingual librispeech Dutch train data.

Zero-shot - 9.3 (MLS Dutch test)
Fine-tune MLS Dutch train - 10.59 (MLS Dutch test)

Even after fine-tuning the model is doing worse than the zero-shot model.

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
training_steps: 4000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2515	0.25	1000	0.2579	12.9776
0.24	0.5	2000	0.2361	11.2418
0.1308	0.75	3000	0.2335	10.7503
0.1072	1.0	4000	0.2031	10.5916

Framework versions

Transformers 4.26.0.dev0
Pytorch 1.13.0+cu117
Datasets 2.7.1.dev0
Tokenizers 0.13.2